Detailed paper information

Back to list

Paper code P57
  1. Andrea Bink University Hospital of Zurich Speaker
  2. Emil Nijhuis University Hospital Basel
  3. Dominic Germanier University Hospital Basel
  4. Severina Leu University Hospital Basel
  5. Sabine Schädelin University Hospital Basel
  6. Anthony De Vere-Tyndall University Hospital Zurich and University of Zurich
  7. Luigi Mariani University Hospital Basel
  8. Christoph Stippich Kliniken Schmieder Allensbach
Form of presentation Poster
  • SSNR-Neuroradiology
Abstract text Aims
Subcompartment analysis has shown prognostic value in imaging of gliomas. While manual volumetry is time consuming and not implemented in clinical routine newer advances in automatic software tools made fully automatic analysis of glioma subcompartments available.

We performed a qualitative and quantitative comparison of fully automatic and visual volumetry in patients with glioblastoma of the sub-compartments necrosis, non-contrast enhancing tumor (NCET), contrast enhancing tumor (CET), edema, total tumor volume (TOTAL) without edema and TOTAL.
Pre-operative MRI of 55 patients with glioblastoma were analyzed fully automatically and visually. Quality of automatic analysis was categorized by four raters in good, partially failed and failed. Statistical analysis was performed by Pearson correlation.

Quality analysis showed in 32 patients good segmentations, partially failed segmentations were found in 19 patients and four failed segmentations (2x movement artifacts, 2x failed scull stripping).
The correlation analysis of fully automatic and visual volumetry resulted in: necrosis rho 0.37 (p=0.05); NCET rho 0.5 (p<0.01); CET rho 0.86 (<0.01); edema rho 0.83 (<0.01); TOTAL rho 0.83 (<0.01); TOTAL with edema rho 0.94 (<0.01). The time needed for automatic evaluation was 5 minutes. Dependent on the tumor complexity visual assessment took between10 to 30 minutes.

Although fully automatic segmentation performed well in unexcelled time, we strongly recommend checking results generated by automatic segmentation by an experienced neuroradiologist due to possible outliers.